Designing the Ontological Foundations for Knowledge Domain Analysis Technology:

An Interim Report Neil BENN*, Simon BUCKINGHAM SHUM, John DOMINGUE, Clara MANCINI Knowledge Media Institute, Centre for Research in Computing, The Open University

Abstract. Research into tools to support both quantitative and qualitative analysis of specialist knowledge domains has been undertaken within the two broadly independent traditions of Bibliometrics and Knowledge Management. The ‘knowledge domain analysis’ (KDA) tools within the first tradition follow a citation-based approach to representing knowledge domains and use citation links as the basis for identifying patterns in the relationships among authors and publications. KDA tools within the second, more recent tradition extend the representational scope to include more features of knowledge domains such as the various types of agents in the domain, their intellectual affiliations, and their research activities, all with the aim of enabling more precise queries about the domains. This second approach depends on the development of software artefacts called ontologies, which are used to explicitly define schemes for representing knowledge domains as well as inference rules to facilitate querying. However, current research into ontologies of scholarly domains has not as yet emphasised the key role of ontologies as vehicles for reuse. This report investigates the design of a reusable KDA ontology, which lays the foundation for future development of KDA tools. Following emerging best practice in the field, the ontology ensures its usability by merging existing ontologies, while also improving its reusability by aligning to generic reference ontology. By characterising knowledge domains as domains of semiotic activity, this report proposes to align the existing KDA ontologies to a generic reference ontology of semiotic components. In addition, this report investigates the use of an upper ontology of coherence connections for defining a core set of inference rules in the final ontology. Furthermore, this report submits that proven network-based analytical techniques from Bibliometrics can be reused to provide the basis for new KDA services. This is demonstrated through applying the ontology to represent and reason about two case study domains. Based on this investigation, this report intends to lay the ontological foundations for new KDA technology research.

Keywords. Scholarly Debate Mapping, Macro-argument Analysis, Ontologies

Introduction

Research into tools to support both quantitative and qualitative analysis of specialist knowledge domains has been undertaken within the two broadly independent traditions of Bibliometrics and Knowledge Management. Knowledge Domain Analysis (KDA) tools within the first tradition (e.g. CiteSeer [1] and CiteSpace [2]) follow a citation-

* Correspondence to: Neil Benn, Knowledge Media Institute, The Open University, MK7 6AA, UK. Tel:

+44 (0) 1908 695837; Fax: +44 (0) 1908 653196; E-mail: n.j.l.benn@open.ac.uk

based approach of representing knowledge domains, where citation links are used as the basis for identifying structural patterns in the relationships among authors and publications. Tools within the second tradition (e.g. Bibster [3], ESKIMO [4], CS AKTIVE SPACE [5], and ClaiMaker [6]) extend the representational approach to include more features of knowledge domains – e.g. the types of agents or actors in the domain, their affiliations, and their research activities – with the aim of enabling more precise questions to be asked of the domain. This second approach depends on the development of software artefacts called ontologies, which are used to explicitly define schemes for representing knowledge domains.

This report describes exploratory research into how these two traditions can be bridged in order to exploit both the benefit of ontologies to enable more feature-rich representations, as well as the established techniques of Bibliometrics for identifying structural patterns in the domain. The first section describes the design of a merged KDA ontology that integrates the existing ontologies specified in [3]- [6] (§1). Next, the paper describes how the merged ontology can be extended to include both a scheme for representing scholarly debates and inference rules for reasoning about the debate (§2). Thirdly, the extended ontology is applied to the representation and analysis of the abortion debate, which demonstrates the benefits of reusing techniques from both traditions (§3). The key lessons from this research are then discussed (§4), before concluding with directions for future research (§5).

1. A merged KDA ontology

One method for merging heterogeneous ontologies requires that the existing ontologies are aligned to a more generic reference ontology that can be used to compare the individual classes in the existing ontologies. Based on the fundamental assumptions that knowledge representation and communication are major activities of knowledge domains, and that knowledge representation and communication constitute semiotic activities, we propose to align the existing ontologies to a reference ontology that describes the interactions between components of any given semiotic activity.

1.1. Invoking a generic theory of semiotics as a reference framework

Semiotics is the study of signs and their use in representation and communication. According to Peirce’s theory of semiotics [7], the basic sign-structure in any instance of representation and communication consists of three components: (1) the sign-vehicle, (2) the object referred to by the sign-vehicle, and (3) the interpretant, which is the mental representation that links the sign-vehicle to the object in the mind of some conceiving agent.

Recent research within the ontology engineering field has introduced a reusable Semiotic Ontology Design Pattern (SemODP) [8] that specifies, with some variation of Peircean terminology, the interactions between components of any given semiotic activity. The SemODP1 is shown in Figure 1.

1 The SemODP pattern shown here is based on an updated version of the ‘home’ ontology from which it is

extracted (accessible at http://www.loa-cnr.it/ontologies/cDnS.owl)

Figure 1. The Semiotic Ontology Design Pattern

In Peircean terminology, the InformationObject class of the SemODP represents

the ‘sign-vehicle’. In the context of knowledge domains, the most typical examples of information objects are publications, which are the main vehicles of knowledge representation and communication. A single publication can be regarded as an Information Object, as can each clause, sentence, table, graph, and figure that is either a verbal or non-verbal expression of knowledge within a publication.

The SemODP classes Description and Entity respectively correspond to the ‘interpretant’ and the ‘object’ in Peircean terminology. The Description class is the abstract, communicable knowledge content that an information object expresses. For example, a single publication is an information object that expresses a thesis (in much the same manner as a novel expresses a particular plot). The Entity class covers any physical or non-physical entity that an Information Object refers to via the ‘is about’ relation.

Finally, the SemODP specifies the Agent class. An agent is required to interpret a given Information Object and in such a case, the agent is said to conceive the Description expressed by that particular Information Object. In knowledge domains, instances of the Agent class include both “agentive physical objects” such as persons and “agentive social objects” such as organisations.

1.2. The core KDA ontology classes

The SemODP is used to merge the existing ontologies through a process of aligning the classes in the existing ontologies to the SemODP classes. The process also reveals the core KDA ontology classes that are based on consensus across the existing ontologies. For example, the consensus classes across the existing ontologies that can play the role of ‘Agent’ are ‘Person’ and ‘Organisation’. However, it should be noted that core classes are not fixed indefinitely and constantly evolve as applications generate experience and consensus changes about what is central [9]. Table 1 shows the core KDA classes and their relationships to SemODP classes as well as existing ontology classes.

Table 1. The SemODP classes, the existing KDA classes they subsume, and the core KDA classes with their main properties.

SemODP class Existing ontology classes2

Core KDA class Properties (Type)

2‘swrc:’ is the prefix for the ontology of the Bibster tool [3]; ‘eskimo:’ is the prefix for the ontology of the

ESKIMO tool [4]; ‘akt:’ is the prefix for the ontology of the CS AKTIVE SPACE tool [5]; ‘scholonto:’ is the prefix for the ontology of the ClaiMaker tool [6].

cdns:Agent swrc:Person, eskimo:Person, akt:Person

mkda:Person name (String)

cdns:Agent swrc:Organisation, eskimo:Organisation, akt:Organisation

mkda:Organisation name (String)

cdns:InformationObject swrc:Publication, eskimo:Publication, akt:Publication

mkda:Publication has Author (Agent); hasTitle (String); expresses (Description)

cdns:Description scholonto:Concept mkda:PropositionalContent verbalExpression (String)

cdns:Description scholonto:Concept mkda:NonPropositionalContent verbalExpression (String)

2. Extending the ontology to support scholarly debate mapping

The work in this section builds on the debate mapping approach of Robert Horn [10] who has produced a classic series of seven debate maps for analysing the history and current status of debate on whether computers can think3. What has emerged from this debate mapping approach is a theory of the structure of debate, which has subsequently been articulated by Yoshimi [11] in what he calls a “logic of debate”. Whereas most argumentation research concentrates on the microstructure of arguments (e.g. the types of inference schemes for inferring conclusions from premises), the concern of a logic of debate is how arguments themselves are “constituents in macro-level dialectical structures” [11]. The basic elements of this logic of debate are implemented as additional classes and relations in the merged KDA ontology.

2.1. Debate representation

Issues In the proposed logic of debate, issues can be characterised as the organisational atoms in structuring scholarly debate4. Indeed, according to [13], one of the essential characteristics of argumentation is that there is an issue to be settled and that the argumentative reasoning is being used to contribute to a settling of the issue. An Issue class is introduced into the ontology as a specialisation of the core KDA class NonPropositionalContent.

Propositions & Arguments The other basic elements in the proposed logic of debate are claims (propositions) and arguments, where the term ‘argument’ is used in the abstract sense of a set of propositions, one of which is a conclusion and the rest of which are premises. However, an argument can play the role of premise in another argument, thus allowing the chaining of arguments. Proposition and Argument classes are introduced into the

3 This debate largely takes place within the knowledge domain of Artificial Intelligence 4 This is analogous to the proposal by [12] that issues are the “organisational atoms” of so-called

information-based information systems for solving intractable design problems.

ontology as specialisations of the core KDA class PropositionalContent. The main relations between claims and arguments in the logic of debate are supports and disputes.

2.2. Debate analysis

Analysing a debate involves reasoning with representations of that debate to detect potentially significant features of the debate. Here we propose to draw on the well-developed network-based analytical techniques employed within the Bibliometrics tradition. However, this reuse is not straightforward since the analytical techniques are typically designed to operate on single-link-type network representations of domains, where the links between nodes are used to signal positive association between nodes. For example, network-based analytical techniques are often applied to co-citation networks where a link between two publications is established when they are both cited by a common third publication, and that link signals positive association between the two publications.

This single-link-type assumption presents a challenge because the ontology-based debate representations can be regarded as ‘multi-faceted’ representations – i.e. there are a number of node types and a number of link types. Thus, before network-based analytical techniques can be reused for debate analysis, transformation rules need to be defined that can project the ‘multi-faceted’ debate representation onto a single-link-type representation.

In order for such a projection to work, the multiple relations in the ontology need to be uniformly interpreted from a single perspective. We propose to interpret the relations in the ontology in a ‘rhetorical-discourse’ context. Indeed, it can be be argued that the analytical techniques of the Bibliometric tradition interpret the publications in a citation-based network as rhetorical viewpoints, which implies that the positive association link between nodes can be interpreted as rhetorical agreement.

Furthermore, the work of Mancini and Buckingham Shum [14] provides the basis for an efficient implementation of the transformation rules. An important feature of their work is the use of a limited set of cognitively grounded parameters (derived from the psycholinguistic work on discourse comprehension 5 by [15]) to define the underlying meaning of discourse relations in the ontology of the ClaiMaker tool. Mancini and Buckingham Shum [14] anticipate that using discourse coherence parameters as the underlying definition language will allow different discourse-relation vocabularies to be used for representing discourse without changing the underlying discourse analysis services provided by their tool. The four bipolar discourse parameters proposed by [15] are: Additive/Causal, Positive/Negative, Semantic/Pragmatic, and Basic/Non-Basic.

The ‘Additive/Causal’ parameter depends, respectively, on whether a weak or strong correlation exists between two discourse units. Note that ‘causal’ is generally given a broad reading in discourse comprehension research to include causality involved in argumentation (where a conclusion is motivated because of a particular line of reasoning), as well as more typical cause-effect relationships between states of affairs.

The ‘Positive/Negative’ parameter depends, respectively, on whether or not the expected connection holds between the two discourse units in question. For example,

5 Discourse comprehension research in general is concerned with the process by which readers are able to

construct a coherent mental representation of the information conveyed by a particular piece of discourse.

in the sentence “Because he had political experience, he was elected president” the connection between the two units is Positive since the reader would typically expect “being elected president” to follow from “having political experience”. However, in the sentence “He did not have any political experience, yet he was elected president” the connection between the two units is Negative since the expected consequent of “not having any political experience” is “not being elected president”, but what is actually expressed is a violation of that expectation – i.e. “yet he was elected president”.

The ‘Semantic/Pragmatic’ parameter depends, respectively, on whether the connection between the two discourse units lies between their factual content or between the speech acts of expressing the two discourse units. At this stage we are primarily focussed on enabling debate analysis, and since debate analysis falls within the realm of speech acts [16], the relations between entities that make up a debate representation will be parameterised as ‘Pragmatic’ by default.

The ‘Basic/Non-Basic’ parameter depends, respectively, on whether or not, in the case of Causal connections, the cause precedes the consequent in the presentation of the discourse. The example – “Because he had political experience, he was elected president” – is parameterised as Basic, whereas the example – “He was elected president because he had political experience” – is parameterised as Non-Basic. This parameter is largely about presentation and does not affect the essential nature or meaning of the discourse connection. Thus it can be omitted from the basic parameterisation of relations in the ontology.

These coherence parameters are then used as a grammar for defining relations in the merged KDA ontology, including relations between publications, between persons, and between arguments. The benefit of this approach is that rather than implement a multitude of inference rules for inferring positive association, only a limited set of parameterised inference rules need to be implemented. For example, Figure 2(i), which shows a parameterised rule for inferring a +ADDITIVE connection between some Y and some Z, covers typical positive association inferences such as when two arguments support a common third argument or when two publications cite a common third publication. Figure 2(ii), which also shows a parameterised rule, covers typical positive association inferences such as when two persons author a common publication or when two arguments are expressed by a common publication. Finally, Figure 2(iii) covers a typical ‘undercutting’ pattern in argument analysis which is a variation of the social network analysis adage that “the enemy of my enemy is my friend”.

Figure 2. Some of the CCR-parameterised inference rules in the ontology. The dotted line indicates that a +ADDITIVE connection is inferred based on the other connections.

3. The abortion debate case study

This section describes the representation and analysis of the debate about the desired legality/illegality of abortions, as laid out in the Abortion Debate entry of the online Wikipedia [17]. This Wikipedia entry provides the ‘raw material’ for capturing computable representations of the debate, thus demonstrating, as a proof-of-concept, the potential use of the ontology in semantically marking up scholarly information resources on the (Semantic) Web.

3.1. Capturing representations of the debate in a knowledge base

Figure 3 (unshaded portion) shows an extract from the Wikipedia entry which expresses some of the debate issues. Figure 3 (shaded portion) then shows how the first of the questions in the extract is captured as an Issue instance in the knowledge base6. As stated previously, the root issue being debated pertains to the legality of abortion. Therefore, an Issue instance (ISS1) is coded in the knowledge base with the verbalExpression attribute assigned the value “What should be the legal status of abortions?”. The Figure then shows the coding of the second Issue instance (ISS2), with the verbalExpression attribute assigned the value “When is the embryo or fetus considered a person?”. Finally, the Figure shows how the subIssueOf relation is used to link the Issue instance ISS2 to the root Issue instance ISS1.

6 The knowledge base is coded using the OCML [18] knowledge modelling language.

Some of the most significant and common issues treated in the abortion debate are:

The beginning of personhood (sometimes phrased ambiguously as "the beginning of life"): When

is the embryo or fetus considered a person? Universal human rights: Is aborting a zygote, embryo, or fetus a violation of human rights? ...

… (def-instance ISS1 Issue ((verbalExpression "What should be the legal status of abortions?"))) (def-instance ISS2 Issue ((verbalExpression "When is the embryo or fetus considered a person?") (subIssueOf ISS1))) … Figure 3. (Unshaded portion) An extract from the Wikipedia entry showing some of the debate issues (Shaded portion) ‘Issue’ instances modelled in the knowledge base.

Next the process turns to representing the claims and arguments in the debate.

According to the Wikipedia entry, the argumentation in the debate is generated by two broadly opposing viewpoints – anti-abortion and pro-abortion. The coding process starts with representing these two basic arguments and then branches off to represent the range of arguments that extend the two broad viewpoints.

Figure 4 (unshaded portion) shows the extract from the Wikipedia entry that gives an overview of the two basic viewpoints. Each viewpoint is expressed in the Wikipedia entry as a set of three premises (the numbered statements). Figure 4 (shaded portion) shows how these two basic viewpoints in the abortion debate are captured in the knowledge base. The three premises for the basic pro-life argument have been captured as Proposition instances P1, P2, and P3. The text just preceding the numbered premises is taken as the conclusion to the premises. The conclusion is represented as another Proposition instance (P4) in the knowledge base. An Argument instance (BASIC-ANTI-ABORTION-ARGUMENT) is then coded, which groups the premises and conclusion together. Similar steps are performed to represent the basic pro-abortion argument in the debate. Finally, the Listing shows the coding of relation instances in the knowledge base. First, an addresses link is established between both of the BASIC-ANTI-ABORTION-ARGUMENT and BASIC-PRO-ABORTION-ARGUMENT Argument instances and the ISS1 Issue instance. Second, a disputes link is asserted, in both directions, between the two Argument instances BASIC-ANTI-ABORTION-ARGUMENT and BASIC-PRO-ABORTION-ARGUMENT.

The view that all or almost all abortion should be illegal generally rests on the claims: (1) that the existence and moral right to life of human beings (human organisms) begins at or near conception-fertilisation; (2) that induced abortion is the deliberate and unjust killing of the fetus in violation of its right to life; and (3) that the law should prohibit unjust violations of the right to life. The view that abortion should in most or all circumstances be legal generally rests on the claims: (1) that women have a right to control what happens in and to their own bodies; (2) that abortion is a just exercise of this right; and (3) that the law should not criminalise just exercises of the right to control one’s own body. Both sides of the debate would grant premise (3) of the central pro-life argument and premises (1) and (3) of the central pro-choice argument. … (def-instance P1 Proposition ((verbalExpression "The existence and moral right to life of human organisms begins at or near conception-fertilisation"))) (def-instance P2 Proposition ((verbalExpression " Induced abortion is the deliberate and unjust killing of the fetus in violation of its right to life")))

(def-instance P3 Proposition

((verbalExpression " The law should prohibit unjust violations of the right to life")))

(def-instance P4 Proposition ((verbalExpression " Abortion should be illegal")))

(def-instance BASIC-ANTI-ABORTION-ARGUMENT Argument ((hasPremise P1 P2 P3) (hasConclusion P4)) (def-instance P5 Proposition ((verbalExpression "Women have a right to control what happens in and to their own bodies")))

(def-instance P6 Proposition ((verbalExpression "Abortion is a just exercise of a woman's right to control what happens in and to her body"))) (def-instance P7 Proposition ((verbalExpression "The law should not criminalise just exercises of the right to control one's own body"))) (def-instance P8 Proposition ((verbalExpression "Abortion should be legal"))) (def-instance BASIC-PRO-CHOICE-ARGUMENT Argument ((hasPremise P5 P6 P7) (hasConclusion P8))) (def-relation-instances (addresses BASIC-ANTI-ABORTION-ARGUMENT ISS1) (addresses BASIC-PRO-ABORTION-ARGUMENT ISS1) (disputes BASIC-ANTI-ABORTION-ARGUMENT BASIC-PRO-ABORTION-ARGUMENT) (disputes BASIC-PRO-ABORTION-ARGUMENT BASIC-ANTI-ABORTION-ARGUMENT)) … Figure 4. (Unshaded portion) An extract from the Wikipedia entry showing the basic anti-abortion and basic pro-abortion viewpoints in the debate. (Shaded portion) Examples of claims and arguments modelled in the knowledge base.

The Wikipedia entry also includes information about all the publications from

which the Wikipedia entry was composed as well as the publication authors who have

participated in the debate. Figure 5 (unshaded portion) shows an extract from the Wikipedia entry detailing the reference information for two publications by the author Judith Thomson, who as participated in the debate. Figure 5 (shaded portion) shows how the author is modelled as a Person instance in the knowledge base (JUDITH_THOMSON) and how the first publication is modelled as a Publication instance (THOMSON1971DEFENSE) with attributes hasAuthor set to the JUDITH_THOMSON Person instance, hasTitle set to "A defense of abortion", and hasYear set to 1971. The shaded portion of the figure also shows how the fact that a publication ‘expresses’ an argument is captured in the knowledge base. In this case, the Judith Thomson-authored publication expresses an argument commonly referred to as the BODILY-RIGHTS-ARGUMENT, and which supports the basic pro-abortion viewpoint in the debate.

Thomson, J. “A Defense of Abortion”. Philosophy and Public Affairs 1:1 (Autumn 1971): 47-66. Thomson, J. "Rights and Deaths". Philosophy and Public Affairs 2:2 (Winter 1973): 146-159.

… (def-instance JUDITH_THOMSON Person) (def-instance THOMSON1971DEFENSE Publication ((hasAuthor JUDITH_THOMSON) (hasTitle "A defense of abortion") (hasYear 1971))) (def-relation-instances (expresses THOMSON1971DEFENSE BODILY-RIGHTS-ARGUMENT) (supports BODILY-RIGHTS-ARGUMENT BASIC-PRO-ABORTION-ARGUMENT)) Figure 5. (Unshaded portion) An extract from the Wikipedia entry showing the reference information for two publications. (ii) (Shaded portion) The corresponding instances modelled in the knowledge base.

3.2. Detecting viewpoint clusters in the debate

The purpose of modelling a debate in a knowledge base is to enable new kinds of analytical services for detecting significant features of the debate. One such feature is the clustering of viewpoints in the debate into cohesive subgroups. A service to detect viewpoint clusters is of significance to a hypothetical end-user because understanding how scholars in a debate fall into different subgroups has already been established as an important part of understanding a debate [10].

The first step in executing this service is to translate the ontology-based representation into a single-link-type network representation so that the clustering technique reused from Bibliometrics can be applied. This involves executing the previously introduced inference rules on the entire ontology-based representation, which results in a network representation with a single +ADDITIVE link type. Figure 6 shows how three of the basic +ADDITIVE inference rules are applied to part of the ontology-based representation of the Abortion debate. The section of the figure labelled (a) shows that a +ADDITIVE relation is inferred between the Argument instances DEPRIVATION-ARGUMENT and ABORTION-BREAST-CANCER-HYPOTHESIS because of common support for BASIC-PRO-LIFE-ARGUMENT. The section labelled (b) shows that the Argument instance BOONIN2003DEFENSE-ARGUMENT is disputing TACIT-CONSENT-OBJECTION-ARGUMENT which in turn is disputing BODILY-RIGHTS-ARGUMENT. The system interprets this as an undercutting move thus it infers a +ADDITIVE relation between BOONIN2003DEFENSE-ARGUMENT and BODILY-RIGHTS-ARGUMENT. The section labelled (c) shows that a +ADDITIVE relation is inferred

between the Argument instances CONTRACEPTION-OBJECTION-ARGUMENT and IDENTITY-OBJECTION-ARGUMENT and EQUALITY-OBJECTION-ARGUMENT because of common dispute of DEPRIVATION-ARGUMENT.

Figure 6. Three of the core +ADDITIVE inference rules being applied to a part of the Abortion debate representation. The link labels are abbreviated as 's' for 'supports', 'd' for 'disputes', and '+A' for '+ADDITIVE'.

Next, a clustering algorithm is run over the network representation, which yields a

number of clustering arrangements, ranging from 2 clusters to 25 clusters. The algorithm designers [19] propose a ‘goodness-of-fit- measure as an objective means of choosing the number of clusters into which the network should be divided. Figure 7 shows the clustering arrangement with five clusters, which the algorithm determines in this case is the arrangement with the ‘best fit’ for the given network data7. However, it should be noted that, as is typical in the Bibliometrics tradition (e.g. [20]), the goal of this type of analysis is typically not to find the definitive clustering arrangement, but rather to detect interesting phenomena that will motivate more focussed investigation on the part of the analyst.

7 The Netdraw network analysis tool (available at http://www.analytictech.com/Netdraw/netdraw.htm) is

used here to perform the clustering and visualisation.

c

a b

Figure 7. Five viewpoint clusters (V.C. #1 – V.C. #5) identified in the abortion debate network. The dashed red lines between clusters indicate opposition (which is determined by further analysis once the clustering results are translated back into the knowledge base)

Once the viewpoint clusters have been identified the results are translated back

into the knowledge base where each of the clusters detected in the network is captured as a ViewpointCluster instance. Further analysis is then performed on these instances to determine, for example, the persons who are associated with each cluster and the clusters which are deemed to be opposing each other. Two clusters are regarded as opposing if at least half of the viewpoints in one cluster have a ‘disputes’ relation with the viewpoints in the other cluster. Table 2 shows for each cluster, the associated viewpoints, the associated person(s), and the opposing cluster(s).

Table 2. Further details of the viewpoint clusters detected in the abortion debate, including the persons associated with each cluster and the clusters that are opposing each other.

VC# Associated Viewpoints Associated Person(s) Opposing Cluster(s)

VC1 COUNTER-NATURAL-CAPACITIES-ARGUMENT8, PERSONHOOD-PROPERTIES-ARGUMENT9

Bonnie Steinbock, David Boonin, Dean Stretton, Jeff McMahan, Louis Pojman, Mary-Anne Warren, Michael Tooley, Peter Singer

VC2 VC4

VC2 COMATOSE-PATIENT-OBJECTION-ARGUMENT10,

Don Marquis, Francis Beckwith, Germaine Grisez, Jeff McMahan, John Finnis, Katherine

VC1 VC4

8 Summary: "The argument that the fetus itself will develop complex mental qualities fails." 9 Summary: "The fetus is not a person because it has at most one of the properties – consciousness – that

characterizes a person." 10 Summary: "Personhood criteria are not a justifiable way to determine right to life since patients in

reversible comas do not exhibit the criteria for personhood yet they still have a right to life"

INFANTICIDE-OBJECTION-ARGUMENT11

Rogers, Massimo Reichlin, Patrick Lee, Robert George, Robert Larmer, Stephen Schwarz

VC3 ABORTION-BREAST-CANCER-HYPOTHESIS12, BASIC-ANTI-ABORTION-ARGUMENT

Don Marquis, Eric Olson, Jim Stone VC4 VC5

VC4 BODILY-RIGHTS-ARGUMENT13, BASIC-PRO-ABORTION-ARGUMENT

David Boonin, Dean Stretton, Jonathan Glover, Judith Thomson, Peter Singer

VC1 VC2 VC3

VC5 CONTRACEPTION-OBJECTION-ARGUMENT14, IDENTITY-OBJECTION-ARGUMENT15

Dean Stretton, Frederick Doepke, Gerald Paske, Jeff McMahan, Lynne Baker, Mary-Anne Warren, Michael Tooley, Peter McInerney, William Hasker

VC3

4. Discussion

The discussion section is organised around a series of questions adapted from the GlobalArgument.net experiment16. These questions were used to evaluate various computer-supported argumentation (CSA) approaches to modelling the Iraq Debate. The discussion is concerned with two main points: the added value of the approach, and the limitations of the approach.

4.1. In what ways does this CSA approach add value?

How does this CSA approach guide a reader/analyst through the debate? The aim of this approach is to provide analytical services that enable the reader to identify interesting features of the debate and gain insights that may not be readily obtained from the raw source material alone. As viewpoint clusters provide a way of abstracting from the complexity of the debate, an approach that enables the detection of viewpoint clusters in a debate is an improvement on what a user would have been able to obtain from looking only at the online Wikipedia entry of the abortion debate. Another interesting feature of the debate that the table reveals is that some persons are associated with more than one cluster. For example, the person of Jeff McMahan is associated with two clusters (VC1 and VC2), and furthermore these clusters happen to oppose each other. These are features of the debate that can then be investigated down to the level of the individual arguments.

11 Summary: "Using personhood criteria would permit not only abortion but infanticide" 12 Summary: "There is a causal relationship between induced abortion and an increased risk of developing

breast cancer" 13 Summary: "Abortion is in some circumstances permissible even if the fetus has a right to life because

even if the fetus has a right to life, it does not have a right to use the pregnant woman's body." 14 Summary: "It is unsound to argue that abortion is wrong because it deprives the fetus of a valuable

future as this entails that contraception, which deprive sperm and ova of a future, is as wrong as murder, something which most people don't believe."

15 Summary: "The fetus does not itself have a future value but has merely the potential to give rise to a different entity, an embodied mind or a person, that would have a future of value"

16 http://kmi.open.ac.uk/projects/GlobalArgument.net

To what extent is the modeller’s or analyst’s expertise critical to achieving the added value? Capturing the debate in the knowledge base often relied on the modeller’s ability to reconstruct argumentation to include parts of arguments not expressed in the original information resource as well as inter-argument relations no expressed in the original information resource. This has an impact on what kinds of connections can be inferred during the reasoning steps, which then has an impact on what features of the debate can be identified.

4.2. What are the limitations of the CSA approach?

What aspects of the debate proved difficult to model? Using this approach it was difficult to account for the different types of disputes relations between arguments. For example, in the case study, one argument often disputed another, not just because of disagreement with the conclusion, but because of the perceived ‘unsoundness’ of the reasoning used to arrive at the conclusion. Also, because of the focus on macro-argumentation, it was difficult to account for different inference schemes for moving from premises to conclusion in individual arguments.

What missing capabilities have been identified? One important missing capability is (semi)automatically capturing debate representations from primary literature sources (e.g. experimental articles carried by scholarly journals). The approach relied on a manual process of constructing representations of the debate based on the Wikipedia, which is classified as a tertiary literature source [21]. Tertiary literature – which includes encyclopaedias, handbooks and review articles – consolidates and synthesises the primary literature thus providing an entry point into the particular domain. This was used for the case study to enable the manual coding of the debate, which would have been too vast to code using all the primary literature that was synthesised. However, this has meant that the debate representations rely on the accuracy of the tertiary-level synthesis of the primary literature.

5. Conclusion and Future Work

This report has described how an ontology of the structure of specialist knowledge domains can be extended to support the representation and analysis of scholarly debates within knowledge domains. The benefit of this extension has then been demonstrated by representing and analysing a case study debate. In particular, a service was demonstrated for detecting ‘viewpoint clusters’ as significant features of scholarly debate. Other debate modelling case studies are currently being written up.

Future directions for this research need to address the difficult question of how to best support the use of primary literature as the source for capturing debate representations. To cover a significant area of any knowledge domain, the modelling of primary literature would need to be conducted in a (semi)automated distributed fashion using possibly many modellers. This would introduce new challenges of trying to ensure consistent modelling across the different modellers. Finally, to address the current limitations of representing micro-argumentation, future research needs to

investigate how the ontology can incorporate the argument specification of the Argument Interchange Format (AIF) [9].

References

[1] Lawrence, S., C. Giles, and K. Bollacker, Digital Libraries and Autonomous Citation Indexing. IEEE Computer, 1999. 32(6): p. 67-71.

[2] Chen, C., CiteSpace II: Detecting and Visualizing Emerging Trends and Transient Patterns in Scientific Literature. Journal of the American Society for Information Science and Technology, 2006. 57(3): p. 359–377.

[3] Haase, P., et al., Bibster - a semantics-based bibliographic Peer-to-Peer system. Journal of Web Semantics, 2004. 2(1): p. 99-103.

[4] Kampa, S., Who are the experts? E-Scholars in the Semantic Web, in Department of Electronics and Computer Science. 2002, University of Southampton.

[5] Schraefel, M.M.C., et al. CS AKTive space: representing computer science in the semantic web. in 13th International Conference on World Wide Web, WWW 2004. 2004. New York, NY, USA: ACM.

[6] Uren, V., et al., Sensemaking tools for understanding research literatures: Design, implementation and user evaluation. International Journal of Human-Computer Studies, 2006. 64(5): p. 420-445.

[7] Atkin, A., Peirce's Theory of Signs, in The Stanford Encyclopedia of Philosophy (Fall 2007 Edition), E.N. Zalta, Editor. 2007, URL = <http://plato.stanford.edu/archives/fall2007/entries/peirce-semiotics/>.

[8] Behrendt, W., et al. Towards an Ontology-Based Distributed Architecture for Paid Content. in The Semantic Web: Research and Applications, 2nd European Semantic Web Conference (ESWC 2005). 2005. Heraklion, Crete, Greece: Springer-Verlag.

[9] Chesñevar, C., et al., Towards an argument interchange format. Knowledge Engineering Review, 2006. 21(4): p. 293-316.

[10] Horn, R.E., Mapping great debates: Can computers think? 7 maps and Handbook. 1998, MacroVU: Bainbridge Island, WA.

[11] Yoshimi, J., Mapping the Structure of Debate. Informal Logic, 2004. 24(1): p. 1-21. [12] Kunz, W. and H.W.J. Rittel, Issues as Elements of Information Systems. 1970, Center for Planning

and Development Research, University of California, Berkeley. [13] Walton, D., Argumentation Schemes for Presumptive Reasoning. 1996, Mahwah, New Jersey:

Lawrence Erlbaum Associates. [14] Mancini, C. and S. Buckingham Shum, Modelling discourse in contested domains: A semiotic and

cognitive framework. International Journal of Human-Computer Studies, 2006. 64: p. 1154-1171. [15] Sanders, T.J.M., W.P.M. Spooren, and L.G.M. Noordman, Towards a taxonomy of coherence

relations. Discourse Processes, 1992. 15: p. 1-35. [16] Eemeren, F.H.v., et al., Reconstructing Argumentative Discourse. 1993, Tuscaloosa/London: The

University of Alabama Press. [17] Website, The Abortion Debate, URL = <http://en.wikipedia.org/wiki/Abortion_debate> (accessed

on 22 September 2006). [18] Motta, E., Reusable Components for Knowledge Modelling: Case Studies in Parametric Design

Problem Solving. Frontiers in Artificial Intelligence and Applications, ed. J.L.d.M. Breuker, R.; Ohsuga, S.; Swartout, W. Vol. 53. 1999, Amsterdam: IOS Press.

[19] Newman, M.E.J., Fast algorithm for detecting community structure in networks. Physical Review E, 2004. 69(066133).

[20] Andrews, J.E., An author co-citation analysis of medical informatics. Journal of the Medical LIbrary Association, 2003. 91(1): p. 47-56.

[21] Anderson, J., The role of subject literature in scholarly communication: An interpretation based on social epistemology. Journal of Documentation, 2002. 58(4): p. 463-481.